Opus Magnum Bench: leaderboard
Puzzle-solving benchmark based on Opus Magnum campaign levels, scoring models by whether they synthesize valid alchemy-machine solutions and how close those solutions are to human-best efficiency.
Metric: Human-normalized score (%). Source: opusmagnumbench.com. Status: years away from saturation. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.6 Sol (xHigh) | 70.86 |
| 2 | Claude Fable 5 (High) | 60.15 |
| 3 | GPT-5.5 (xHigh) | 53.56 |
| 4 | Claude Fable 5 (Low) | 49.28 |
| 5 | GPT-5.6 Terra (xHigh) | 47.72 |
| 6 | GPT-5.5 (Medium) | 37.79 |
| 7 | Gemini 3.1 Pro (Preview) (High) | 31.11 |
| 8 | Claude Sonnet 5 (High) | 26.55 |
| 9 | GLM-5.2 (High) | 25.39 |
| 10 | Gemini 3.5 Flash (High) | 24.49 |
| 11 | Claude Opus 4.8 (High) | 22.81 |
| 12 | Grok 4.5 (High) | 18.19 |
| 13 | GPT-5.6 Luna (xHigh) | 17.8 |
| 14 | Gemini 3 Flash (High) | 15.39 |
| 15 | DeepSeek V4 Pro | 13.39 |
Interactive version: theaggregate.ai/benchmark?slug=opus-magnum-bench · How It Works · Data refreshed daily, snapshot 2026-09-05.